Download - Testing in Service-Oriented Environments
-
8/9/2019 Testing in Service-Oriented Environments
1/79
Testing in Service-Oriented Environments
Ed Morris
William Anderson
Sriram Bala
David Carney
John Morley
Patrick Place
Soumya Simanta
March 2010
TECHNICAL REPORTCMU/SEI-2010-TR-011ESC-TR-2010-011
Research, Technology, and System Solutions (RTSS) ProgramUnlimited distribution subject to the copyright.
http://www.sei.cmu.edu
http://www.sei.cmu.edu/http://www.sei.cmu.edu/ -
8/9/2019 Testing in Service-Oriented Environments
2/79
This report was prepared for the
SEI Administrative AgentESC/XPK
5 Eglin Street
Hanscom AFB, MA 01731-2100
The ideas and findings in this report should not be construed as an official DoD position. It is published in the
interest of scientific and technical information exchange.
This work is sponsored by the U.S. Department of Defense. The Software Engineering Institute is a federally
funded research and development center sponsored by the U.S. Department of Defense.
Copyright 2010 Carnegie Mellon University.
NO WARRANTY
THIS CARNEGIE MELLON UNIVERSITY AND SOFTWARE ENGINEERING INSTITUTE MATERIAL IS
FURNISHED ON AN "AS-IS" BASIS. CARNEGIE MELLON UNIVERSITY MAKES NO WARRANTIES OF
ANY KIND, EITHER EXPRESSED OR IMPLIED, AS TO ANY MATTER INCLUDING, BUT NOT LIMITED
TO, WARRANTY OF FITNESS FOR PURPOSE OR MERCHANTABILITY, EXCLUSIVITY, OR RESULTS
OBTAINED FROM USE OF THE MATERIAL. CARNEGIE MELLON UNIVERSITY DOES NOT MAKE
ANY WARRANTY OF ANY KIND WITH RESPECT TO FREEDOM FROM PATENT, TRADEMARK, OR
COPYRIGHT INFRINGEMENT.
Use of any trademarks in this report is not intended in any way to infringe on the rights of the trademark holder.
Internal use. Permission to reproduce this document and to prepare derivative works from this document for
internal use is granted, provided the copyright and "No Warranty" statements are included with all reproductions
and derivative works.
External use. This document may be reproduced in its entirety, without modification, and freely distributed in
written or electronic form without requesting formal permission. Permission is required for any other external
and/or commercial use. Requests for permission should be directed to the Software Engineering Institute at
This work was created in the performance of Federal Government Contract Number FA8721-05-C-0003 with
Carnegie Mellon University for the operation of the Software Engineering Institute, a federally funded research
and development center. The Government of the United States has a royalty-free government-purpose license to
use, duplicate, or disclose the work, in whole or in part and in any manner, and to have or permit others to do so,
for government purposes pursuant to the copyright license under the clause at 252.227-7013.
mailto:[email protected]:[email protected] -
8/9/2019 Testing in Service-Oriented Environments
3/79
i | CMU/SEI-2010-TR-011
Table of Contents
Abstract vii1 Introduction 1
1.1 Organization of This Report 22 Testing in a Service Oriented System 3
2.1 Artifacts to be Tested 42.2 Perspectives, Roles, and Responsibilities 6
3 Testing Challenges 93.1 SOA Infrastructure Testing Challenges 93.2 Service Challenges 103.3 Environment for Testing 12
4 Testing Functionality 134.1 Testing SOA Infrastructure Capabilities 134.2 Testing Web Service Functionality 144.3 Fault Injection 154.4 Regression Testing 154.5 Testing Business Processes That Span Traditional Boundaries 17
4.5.1 Process for End-to-End Testing 184.5.2 Defining End-to-End Threads and Establishing Test Scenarios 194.5.3 Identifying Dynamic Mission Threads 204.5.4 Identifying and Initializing the Context and Test Payloads 214.5.5 Capturing and Analyzing Results 21
5 Testing For Interoperability 235.1 Defining Interoperability Levels 235.2
Interoperability in Infrastructure 25
5.2.1 Testing Within the Infrastructure 255.2.2 Testing Between Infrastructures 25
5.3 Interoperability in Services 265.4 End-to-End Interoperability 26
6 Testing for Security 296.1 Thread Modeling and Attack Surface 29
6.1.1 Threat Modeling 296.1.2 Attack Surface 306.1.3 Combining Threat Modeling with Attack Surface 306.1.4 Use of Threat Modeling with Attack Surface in SOA Testing 31
6.2 Testing SOA Infrastructure Security 316.3 Testing Web Service Security 326.4 Web Service Verification 33
7 Testing for Other Quality Attributes 357.1 Performance 357.2 Reliability 37
8 Testing for Standards Conformance 399 Test-Related Strategies 41
-
8/9/2019 Testing in Service-Oriented Environments
4/79
ii | CMU/SEI-2010-TR-011
9.1 Test-Driven Development 419.2 Design-by-Contract 429.3 Governance Enforcement 429.4 Runtime Monitoring 439.5 Service Level Agreements 449.6 Assurance Cases 45
10 Summary 47Appendix A List of Acronyms Used 49Appendix B Consolidated List of Recommendations 51Appendix C Key Attributes for Testing Web Services 55Appendix D Testing Process Preconditions and Deliverables 59References 63
-
8/9/2019 Testing in Service-Oriented Environments
5/79
iii | CMU/SEI-2010-TR-011
List of Figures
Figure 1: Mission Thread Testing Process 19Figure 2: SOA Interoperation Stack Showing Levels of Interoperability 23Figure 3: Threats and Attack Surfaces 31Figure 4: Overview of the Web Service Verification Process 34Figure 5: Potential Performance Bottleneck Points in a Web Service Invocation 36Figure 6: Enterprise Service Management (ESM) in an SOA Environment 44
-
8/9/2019 Testing in Service-Oriented Environments
6/79
iv | CMU/SEI-2010-TR-011
-
8/9/2019 Testing in Service-Oriented Environments
7/79
v | CMU/SEI-2010-TR-011
List of Tables
Table 1: Implications of Service-Orientation on Quality Attributes 3Table 2: Important Roles in an SOA Environment 6Table 3: Recommendations for Testing Selected Infrastructural Capabilities 13Table 4: Types of Threats and Typical Countermeasures 30Table 5: Performance Metrics 36
-
8/9/2019 Testing in Service-Oriented Environments
8/79
vi | CMU/SEI-2010-TR-011
-
8/9/2019 Testing in Service-Oriented Environments
9/79
vii | CMU/SEI-2010-TR-011
Abstract
This report makes recommendations for testing service-oriented architecture (SOA) implementa-
tions consisting of infrastructure, services, and end-to-end processes. Testing implementations of
SOA infrastructure, services, and end-to-end processing in support of business processes is com-
plex. SOA infrastructure is often composed of multiple, independently constructed commercial
productsoften from different vendorsthat must be carefully configured to interact in an ap-
propriate manner. Services are loosely coupled components which are intended to make minimal
assumptions about where, why, and under what environmental conditions they are invoked. Busi-
ness processes link together multiple services and other systems in support of specific tasks.
These services and systems may operate on remote platforms controlled by different organizations
and with different SOA infrastructures. Such complications make it difficult to establish appropri-
ate environments for tests, to ensure specific qualities of service, and to keep testing up-to-date
with changing configurations of platforms, infrastructure, services, and other components.
-
8/9/2019 Testing in Service-Oriented Environments
10/79
viii | CMU/SEI-2010-TR-011
-
8/9/2019 Testing in Service-Oriented Environments
11/79
1 | CMU/SEI-2010-TR-011
1 Introduction
Service-oriented architecture (SOA) has quickly become a dominant paradigm for distributed,
interoperating, software systems of systems.
1
Three interrelated concepts need to be distinguished:SOA, services, and business processes.
1. SOA is a way of designing, developing, deploying, and managing systems that is characte-
rized by
a. coarse-grained services that represent reusable business functionality
b. service consumers that are clients for the functionality provided by the services, such as
end-user applications, internal and external systems, portals, or even other services (i.e.,
composite services) and that compose applications or systems using the functionality
provided by these services through standard interfaces
c. an SOA infrastructure that provides means to connect service consumers to services [1].
2. Services are reusable components that represent business tasks, such as customer lookup,weather, account lookup, or credit card validation. Services represent a task or step in a busi-
ness process and they can be globally distributed across organizations and reconfigured to
support new business processes.
3. Business processes (also referred to as mission threads by some government organizations)
are a defined set oftasks that produce specific capabilities or products. Each task is com-
posed of one or more services that are integrated to form the process.
While there are many potential implementations of the SOA architectural style, web services that
include HTTP, SOAP, WSDL, and UDDI provide one of the most widely adopted approaches to
SOA implementation [2,3,4,5].2
Because web service implementations are so widespread, com-
mercial tool support is available for many aspects of the engineering process, including testing.SOA implementations are possible using other technologies and standards; however, commercial
tool support is less widely available for these other implementations.
Whichever technologies and standards SOA implementations use, it is challenging to test those
implementations, for various reasons. For instance, services are often outside the control of the
organization consuming the service (i.e., service builders, service owners, and service consumers
may each be from different organizations), leading to potential mismatches and misunderstand-
ings between parties. In addition, SOA implementations are often highly dynamic, with frequently
changing services and service consumers, and varying load on services, SOA infrastructure, and
the underlying network. Consequently, it is normally very difficult to replicate all possible confi-
gurations and loads during the testing process.
The scope of an SOA testing effort includes which aspects should be tested for (or against), the
elements to be tested, and artifacts produced during the testing effort. Typically, testing hand-
1 Maier [61] defines a system of systems as one in which there is managerial independence, operational inde-
pendence, and geographic distribution of the elements; continuing evolution; and emergent behaviors such that
the system of systems exhibits behaviors apart from those of individual elements
2Acronyms used in this report are identified in Appendix A.
http://en.wikipedia.org/wiki/Taskshttp://en.wikipedia.org/wiki/Tasks -
8/9/2019 Testing in Service-Oriented Environments
12/79
2 | CMU/SEI-2010-TR-011
books include strategies for testing the completeness, correctness, and accuracy of requirements,
architecture, design, documentation, and many other non-executable artifacts. This report does not
address testing of non-executable artifacts. Here, we focus on dynamic testing [6], which includes
unit, integration, and systems-level testing of executable software (for both functional and non-
functional attributes). For information about the broader view of testing, see Kaner, Falk, and
Nguyen, and Whittaker [6,7].
1.1 Organization of This Report
Section 2 introduces foundational material about SOA implementation testing. Three aspects to be
tested are described: functionality, non-functional attributes, and conformance. In addition, the
elements to be testedSOA infrastructure, web services, and end-to-end threadsare discussed.
Finally, key roles such as service developer, service provider, and service consumer are examined.
Section 3 details the challenges faced when testing SOA implementations; Sections 48 provide
specific guidance for testing the aspects of the implementation. Section 9 discusses other tech-
niques that can be used in conjunction with traditional testing in order to enhance the testing
process. Section 10 is a short summary of the report.
Throughout the report, we highlight recommendations for good SOA implementation testing prac-
tice; the recommendations are compiled in Appendix B.
-
8/9/2019 Testing in Service-Oriented Environments
13/79
3 | CMU/SEI-2010-TR-011
2 Testing in a Service Oriented System
At the highest level, testing in an SOA implementation does not differ from testing in traditional
systems. Specifically, testing must addressfunctionality: The functional capabilities must be tested to determine whether requirements
and other expectations will be (or are being) met. For example, functional testing in the SOA
infrastructure can verify that services are made available correctly (i.e., they are published),
that the services can be located, and that connections can be established (calledfind and
bind).
non-functional attributes: The non-functional characteristics that are essential for determin-
ing fitness for operational use, also called quality attributes, must be tested. Table 1 is a list
of some quality attributes [8] considered important, with implications from service-
orientation on testing.
conformance: Testing can verify conformance to specific guidelines and standards that theorganization has chosen to adopt. For implementations based on web services, there are
many standards and guidelines that can be found from the W3C [9] the WS-I [10], and
OASIS [11].3
In addition, many organizations develop standards and guidelines specific to
their lines of business.
Table 1: Implications of Service-Orientation on Quality Attributes
Quality Attribute Implications of Service Orientation on the Quality Attribute
Availability A service may not be under the direct control of the service consumer. Therefore,
there is no guarantee of availability.
Multiple identical instances of a service may be used to enhance availability, lead-
ing to a need to test consistency between the instances.
Diverse instances of a service may be used, complicating testing as each imple-mentation must be tested.
Modifiability A service-oriented system is meant to be easy to modify. Further, because its in-
ternal implementation (e.g., program language, structure) is hidden from service
consumers, a service implementation can be changed as long as the service inter-
face is unchanged. The interface can also be modified as long as prior functionality
is maintained (i.e., functionality can be added incrementally). Testing is compli-
cated in cases where the testers do not have access to service implementations. In
such a case, testing becomes similar to a COTS (commercial off-the-shelf) envi-
ronment (i.e., black box with significant trust issues)
A new version of a service may not be backwards compatible with earlier, tested
versions of the service.
Interoperability Services are meant to be interoperable by publishing their service interfaces and
using common protocols, however, there is minimum support for semantic intero-
perability.SOA infrastructures can differ in the way they implement features (e.g., security).
This makes it difficult to invoke services across differing infrastructures. This prob-
lem is sometimes referred to as infrastructure federation.
3Acronyms used in this report are identified in Appendix A.
-
8/9/2019 Testing in Service-Oriented Environments
14/79
4 | CMU/SEI-2010-TR-011
Quality Attribute Implications of Service Orientation on the Quality Attribute
Performance If implemented using web services, service orientation has a negative effect on run-
time performance (i.e., response time and throughput) primarily because
XML serialization, deserialization, validation, and transformation produce per-
formance overhead.
The presence of meta information makes web services messages more verbose
and hence larger in size; therefore, the messages need more bandwidth.
Adaptability Services are intended to be rapidly adaptable to changing business or mission de-
mands. Because the environment is changing rapidly, testing must balance between
the desire for full testing and the need for rapid deployment. Automated test suites
and regression testing techniques become critical.
Reliability While standards are available to help with message-level reliability, service-level
reliability is still implementation specific. Tests need to be developed for the failures
that derive from the distributed nature of an SOA-based system. For example, testing
should analyze the behavior when an invoked service is unavailable.
Security Service consumers can reuse existing services; however, this advantage often comes
at a price. An end-to-end thread can cross trust boundaries. Testing must validate
that security is maintained even when information flows across untrusted networks
[12].
2.1 Artifacts to be Tested
The functional, non-functional, and conformance characteristics to be tested should be considered
in the context of the artifacts of a service-oriented system that are available for testing, specifical-
ly the SOA infrastructure, the services, and the service consumers. Note that service consumers
may be services that invoke other services, or they may be composite services, or even end-to-end
threads performing an enterprise-level capability.
SOA Infrastructure
The SOA infrastructure typically consists of capabilities and services to register and discover ser-
vices, manage metadata, provide security, and deliver messages. The infrastructure may be com-posed of custom-developed software, commercial products, or some combination of custom,
commercial, and open source capability. An SOA infrastructure stack typically consists of at least
the following
a registry
an enterprise service bus (ESB)
application servers
web servers
data stores and databases
However, there is wide variation in the SOA infrastructure capabilities provided by commercialvendors, in terms of (1) core elements provided by the infrastructure stack; (2) tools to support
developing, integrating, testing, and other software engineering activities; and (3) capabilities,
such as a repository or support for security (e.g., services for authentication or authorization of
access).
-
8/9/2019 Testing in Service-Oriented Environments
15/79
5 | CMU/SEI-2010-TR-011
Web Services
Testing of web services includes testing individual services and composites of services. An indi-
vidual web service (also called an atomic web service) typically provides coarse-grained, busi-
ness-level capability. The elements of a web service from a testing perspective are the service in-
terface, service implementation, message format, message payload, and service level agreement
(SLA).
Composites involve multiple web services that interact to provide a capability. Web service or-
chestration and web service choreography are two types of composites [13,14]commonly found in
SOA environments.
Web service orchestration directs the actions of one or more individual web services such
that they work together to perform some useful function. Important elements of a web ser-
vice orchestration are the composition code (e.g., WS-BPEL code), the individual web ser-
vices, and the infrastructure elements and capabilities used to realize the orchestration.
Orchestration (quoting from aBPTrends online column) [15]:
Defines a single master controls of all aspects of a process (top-down approach)Supports a graphical view of the sequence
Easily maps to SOA
Is usually simpler to start with; but often harder to scale to more complex processes
Is driven by the graphical sequence model, i.e. function follows form
Represents the state-of-the-practice, and is supported by the majority of tools
Web service choreography involves cooperating web service participants, in which services
act as peers involved in interactions that may be long-lived and rely on state information to
direct behavior (i.e., stateful interactions). The goal of these interactions is common and
complementary behavior of the participants that cumulatively accomplishes a business goal.
Choreography (quoting from anBPTrends online column) [15]:
The overall process behavior emerges from the working of its parts (bottom up). No
global perspective is required
Complex work processes are decomposed into work agendas where each autonomous
element controls its own agenda
Easily maps to event and agent based systems
Is usually more difficult to start, but often easier to scale to complex processes
Graphical representations can be derived from the process, i.e. form follows function
Represents the state-of-the-art, and is gaining support with emerging tools
End-to-End Threads
End-to-end threads are the combination of humans, applications, services, back-end applications,
and databases that utilize the SOA and network infrastructure to perform a business task.4
End-to-
end threads include services along with other interacting components (human, functional, infra-
structural), along with the operating environment.
4Along with end-to-end threads, SOA-based applications are tested in this aspect. For simplicity, we use end-to-end thread (or threads) to stand for both types of composites.
http://en.wikipedia.org/wiki/Web_Servicehttp://en.wikipedia.org/wiki/Web_Service -
8/9/2019 Testing in Service-Oriented Environments
16/79
6 | CMU/SEI-2010-TR-011
Recommendation for SOA Testing
Rec. 1: Develop an SOA test strategy that accounts for SOA infrastructure elements, web services
(individual and composites), and end-to-end threads.
2.2 Perspectives, Roles, and Responsibilities
As indicated in Table 2,5
there are several roles that organizations and individuals play in devel-
oping an SOA implementation. Each role is important in SOA testing, and failure to define the
role-specific expectations regarding testing can lead to problems in testing and issue resolution.
For example, it is important to determine the role of a COTS vendor in testing and resolving is-
sues with its product in the users context. Clear expectations can help avoid the finger-pointing
that sometimes occurs when products do not interact as expected or promised. Likewise, defining
the role of individual service providers in relation to testing of composites (orchestrations and
choreographies) or end-to-end threads simplifies the processes of setting up the environment, ex-
ecuting tests, and analyzing results.
Table 2: Important Roles in an SOA Environment
Role Action Recommendation for SOA Testing
Service developer Creates the interface of an individual
service and its underlying implementa-
tion by using an existing component and
wrapping it as a service or forming the
service implementation fromscratch
Rec. 2: Establish guidelines for testing the
service before it is ready for use either
by the service consumers or the ser-
vice integrator.
Service provider Provides services. A service provider
may or may not be the developer of the
service; however, a service developer
can also be a service provider.
Rec. 3: Establish clear lines of testing and
customer support responsibility that al-
low for distinctions between the roles
of service developers and service pro-
viders.
Service consumer Uses a service (individual or composite)
directly
Rec. 4: Develop governance processes that
support service consumers in the iden-tification of use cases and the tests
necessary to achieve appropriate as-
surance about performance of the ser-
vice within those use cases.
Service integrator Uses existing services (individual or
composite) either to create composite
services or to create an end user appli-
cation
Rec. 5: Develop guidelines for testing compo-
sites of various types, including com-
posites that implement capabilities
supporting pre-defined mission threads
and ad hoc composites that reflect at-
tempts to address emerging needs.
Rec. 6: Develop guidelines for testing compo-
sites employing the range of composi-tion mechanisms expected (e.g., WS-
BPEL defined, application-embedded
composites, WS-CDL).
5The table is an extension of roles identified in Testing services and service-centric systems: challenges andopportunities [66].
-
8/9/2019 Testing in Service-Oriented Environments
17/79
7 | CMU/SEI-2010-TR-011
Role Action Recommendation for SOA Testing
Infrastructure
provider
Provides the necessary SOA infrastruc-
ture middleware (e.g., ESB) and infra-
structural mechanisms such as service
discovery to service providers, service
consumers, and service integrators
Rec. 7: Develop guidelines and governance
processes for testing/verification of
new and revised infrastructure capa-
bilities, including notification of users
of infrastructure changes and trig-
gers for retesting.
Rec. 8: Develop policies for the type and
level of testing support provided by
the infrastructure provider to the ser-
vice provider or integrator.
Third-party service
tester or certifier
Validates and potentially certifies wheth-
er a service (individual or composite)
works as expected
Rec. 9: Identify the focus, expectations, and
limitations of third-party test-
ing/certification activities.
End user Uses applications that employ services Rec. 10: Develop policy clearly delineating
the forms of testing performed ac-
cording to SOA roles.
-
8/9/2019 Testing in Service-Oriented Environments
18/79
8 | CMU/SEI-2010-TR-011
-
8/9/2019 Testing in Service-Oriented Environments
19/79
9 | CMU/SEI-2010-TR-011
3 Testing Challenges
Testing a service-oriented system is made challenging by a number of factors, not least of which
is that most such systems are not only componentized, but also distributed. In addition to the tra-
ditional problems of distributed systems such as communication failure, the nature of service
orientation, where different components may be provided by different organizations leads to addi-
tional challenges. We have already identified the various structural components involved and now
discuss the challenges that each presents.
3.1 SOA Infrastructure Testing Challenges
We have identified five factors that complicate testing of SOA infrastructure: limited technical
information about components, complex configuration of an infrastructure, rapid release cycles
for components in an infrastructure, lack of a uniform view, and variation across an infrastructure.
The infrastructure provider rarely has access to the design information and source code that
is desired for systems with strict quality requirements (e.g., high performance, security, safe-
ty, etc.). While systems with such requirements may choose components from a collection of
previously accredited systems, even systems with less-stringent quality requirements will
face the challenge of testing black box components.
An SOA infrastructure consists of a number of software components such as registry, reposi-
tory, ESB, and databases. Typically, most of these are COTS components, rather than cus-
tom components developed to satisfy specific functional or quality requirements. Whenever
the SOA infrastructure is composed of components from different vendorsand even when
it is provided by a single vendorindividual components have unique installation, adminis-
tration, and configuration requirements. Each component must be configured correctly for
testing, and testing results can vary across multiple configurations of the same infrastructurecomponents.
There may also be multiple patches and service packs available for individual components
which must be aligned. For example, a quick search performed for this report identified sev-
en different software releases (versions, service packs, vulnerability patches) for one com-
mon ESB during the 13-month period from January 1, 2008 through January 31, 2009. This
rapid cycle of updates necessitates an infrastructure testing process that is fast and efficient.
Many individual components in the SOA infrastructure have their own applications for man-
aging, data recording, and optimizing the component. A single unified view that supports the
tester is often not available.
Implementations of web services stacks (the infrastructure that supports various web servicesstandards) can be inconsistent regarding the COTS, government off-the-shelf (GOTS), open
source, and other products employed; the subsets and versions of SOA standards imple-
mented; and other characteristics. Because COTS products are black boxes, such inconsis-
tencies can complicate testing by introducing factors, such as different error handling me-
thods and inconsistent use of standards that are not understood clearly by developers and
testers.
-
8/9/2019 Testing in Service-Oriented Environments
20/79
10 | CMU/SEI-2010-TR-011
3.2 Service Challenges
The following are key testing challenges from an individual (atomic) web service perspective.
unknown contexts and environments: One of the biggest advantages of web services is their
reusability in variety of contexts. From a service developers perspective (that is, having re-
sponsibility for unit testing the service), it is not easy to anticipate all situations where a ser-vice may be used. This lack of knowledge about usage contexts hampers testing, although
being aware of all contexts is not necessarily a solution either. To address this problem, it is
necessary to determine a set of contexts (based on best estimates) and prepare tests in these
contexts.
unanticipated demand and impact on quality of service: Unlike traditional software compo-
nents, a single instance of a web service can be used by multiple consumers. Since the spe-
cific usage of the service (e.g., load, network and infrastructure delay, data) is unknown at
the time of development and deployment, it is difficult to test and verify whether a service
can successfully meet the quality of service (QoS) expectations of users. Doing so requires
testing for performance under different conditions, such as varying loads.
lack of source and binary code: In many cases, the source code and binary code [16] of the
service and its underlying components are unavailable to the service integrator and tester. In
the absence of source and object code, white box testing techniques such as static analysis
are impossible.6
This circumstance is particularly problematic for organizations that maintain
high information assurance standards, since it becomes difficult to verify that a service is not
acting in a dangerous manner. In most cases, these loosely coupled services are black boxes.
Design and code-level information is often not be available to service providers, meaning
that subtle assumptions embedded in the code can only be identified through rigorous test-
ing.
standards conformance: Web services are based on standards that facilitate syntactic intero-
perability. These standards are defined at a high level of abstraction and are required to con-
form to other complementary standards [17]. This level of abstraction leads to significant
variation in actual vendor implementations, creating additional work for the service tester to
verify compliance with standards such as SOAP, REST [18] , XML, UDDI, WSDL, and
HTTP.7
Tools support testing of many of these well-established standards and they should be
used as appropriate. However, tool support is lacking for several recommendations and stan-
dards that are newer and less widely adopted (e.g., WS-Notification [19], WS-Trust [20]).
In addition to the challenges posed by testing individual services, the following challenges ac-
company testing composite services and end-to-end threads:
service availability: Composites use services that may be controlled by multiple service pro-
viders, and testing of a composition requires that the correct versions of all invoked services
are available. Therefore, the schedule for testing the composite is strongly dependent on
schedules for other services outside the control of the testing organization.
6 White box or glass box testing usually is performed with the knowledge of the internals of an SOA element.
7Acronyms used in this report are identified in Appendix A.
-
8/9/2019 Testing in Service-Oriented Environments
21/79
11 | CMU/SEI-2010-TR-011
multiple provider coordination: Even when a service integrator has access to all participating
services and components, the individual services may be black boxes to the integrator. Thus,
the service integrator may need to communicate with the teams providing these services in
order to learn about the implementation characteristics of the service. For example, if a par-
ticipating service saves information in a database that is not visible to the integrator, the in-
tegrator will need to work with the service provider in order to get access to that information.While access to such information may not always be necessary, it will be a common enough
occurrence that the testing will likely encounter need like this in any significant system.
common semantic data model: There will be a large potential for misunderstandings of the
data exchanged unless all services participating in a composite agree on a common semantic
data model. However, achieving consensus on common semantics is a notoriously difficult
job, and its critical for the tester to determine that the applications create data consistent
with the data model and that they transform it according to specified rules. Inconsistencies in
how data is understood can be very subtle and show up only in a few situations; as a result,
they may be difficult to identify. For example, consider that two services may use tempera-
ture but one measures in Celsius and the other in Fahrenheit.lack of a common fault model: Given that services will likely be developed by different ser-
vice developers, for different purposes, it is unlikely that they will subscribe to a common
fault model. Reconciling fault models is as hard as achieving a common data model and
leads to difficulties when testing that service faults are handled appropriately.
transaction management: Different service providers and SOA infrastructures may imple-
ment slightly different transaction management approaches (e.g., they may have different
models of when a change to a data item actually occurs). These differences may only show
up in unusual or race conditions where the split-second timing of multiple transactions leads
to inconsistent states or service failure. Therefore, the correctness and consistency of all
transactions must be tested.
side effects: Because a single service can be part of multiple composites, execution of test
cases for one composite may produce unwanted side effects in others [21] employing the
service. Also, changes in data or state caused by testing can interfere with other service com-
posites or produce unwanted changes in them. Of course, not all services will suffer from in-
terference between threads; indeed, it is desirable to develop services without such interfe-
rence. However, the difficulty for the tester is determining that any given service is free of
side effects.
independent evolution of participating services: The service integrator often has no control
over individual services or their evolution, yet any change in a participating service can re-
sult in a failure of the composite. If it is unaware of service changes, the service integrator
may be unable to identify the source of a failure. Even when it is aware of the changes, the
service integrator still needs to perform a set of regression tests to ensure the proper func-
tioning of the composite.
lack of tool support: Current testing tools and technologies that claim support for composites
often provide the ability to validate whether a business process representation (e.g., WS-
BPEL) is well-formed and check for the existence of the endpoints of the individual web
-
8/9/2019 Testing in Service-Oriented Environments
22/79
12 | CMU/SEI-2010-TR-011
services that make up the composite web service. Tool support for more comprehensive test-
ing of composites is an active research area.
3.3 Environment for Testing
A final challenge, but one that is necessary to overcome, is the creation of a testing environment
that is as similar as possible to the deployment environment (a high fidelity copy) that allows
developers to test and integrate services and service orchestrations prior to actual deploy-
ment
testers to design, implement, execute, and manage the testing process,8
and to capture infor-
mation about the functional and quality aspects of SOA services, infrastructures, and end-to-
end threads
Establishing a testing environment is difficult because many deployment environments are widely
distributed, with heterogeneous hardware and software (SOA infrastructure, services, operating
systems, and databases). Because of cost and security, it may not be possible for all organizations
hosting a capability in the deployed environment to mirror those capabilities in a testing environ-
ment.
An alternative involves developing a testing environment that mirrors several key aspects of the
deployment environment and provides workarounds for missing aspects. In some cases, test cop-
ies of deployment infrastructure may be available; in other cases, virtual machines can be stood
up to mirror parts of the infrastructure. This testing environment should be available to developers
for early testing of services and service orchestrations, as well as to testing organizations perform-
ing certification activities. The testing environment should be populated with
a suite of testing tools to manage tests, validate conformance to standards, and analyze run-
time functionality and performance
test data and associated databases sufficient to support the execution of testsa mechanism to determine the testing/certification status of a service or service orchestration
(end-to-end thread)
versions of deployed infrastructure and services sufficient to support the execution and test-
ing of services/service orchestrations under development and test
a mechanism for stubbing out or simulating services that are not yet developed
Recommendation for SOA Testing
Rec. 11: Provide developers with a fully outfitted test environment for early testing of services andservice orchestrations; provide this environment to testing organizations, too, for performing
certification activities.
8While this report is not about defining a process for testing SOA implementations, we have included somethoughts about testing process inputs and outputs in Appendix D.
-
8/9/2019 Testing in Service-Oriented Environments
23/79
13 | CMU/SEI-2010-TR-011
4 Testing Functionality
4.1 Testing SOA Infrastructure Capabilities
An SOA infrastructure provides capabilities, often in the form of common services that can be
used by a number of business processes. Services are usually composed of COTS components that
are tested by their vendors, which may mask key information (such as the test cases or their con-
text) from the infrastructure provider. Recommendations for testing SOA infrastructure capabili-
ties are shown in Table 3.
Table 3: Recommendations for Testing Selected Infrastructural Capabilities
Infrastructural
Capability
Implication for SOA Testing Recommendation for SOA Testing
Service Registration
Mechanism
The goal of testing service registration is to
ensure that only validated services become part
of the SOA registry. A final testing results report
is sent to the service provider once the auto-
mated testing is complete. It is also possible for
the service providers to submit test cases con-
forming to a standard when they submit a ser-
vice for registration.
Testing the service registration mechanism
involves testing the tools and options provided
by the infrastructure to register a new service. It
also needs to take into account the version
control of services already registered.
Rec. 12: Develop strategies for testing of the
service registration mechanism.
Service and Resource
Discovery Mechanism
Testing the service and resource discovery
mechanism ensures that registered services
and infrastructure resources can be found byauthorized service consumers. Testing also
needs to be performed if the service consumers
can search for the services they require using
various criteria and matching rules.9
Rec. 13: Develop tests to determine whether
authorized service consumers can
find services based on syntactic orsemantic approaches.
Subscription and Noti-
fication Mechanisms
If the infrastructure provides subscription me-
chanisms, then these need to be tested. All
service consumers must be informed when a
new version of a service is registered with the
infrastructure. Service providers and consumers
also require notifications of service level viola-
tions.
Rec. 14: Develop tests that reveal whether
all appropriate service providers
and consumers are notified when a
new version of a service is regis-
tered with the infrastructure or
when service levels are violated.
9A discovery mechanism can be syntactic or semantic. Syntactic discovery, while easy to implement and test,has limitations because service consumers may not always be aware of the keywords to use in a search, andthe use of just keywords to describe services often ignores their operational characteristics. A semantic ap-proach promises a greater probability that services sought will be found, but requires more testing effort on thepart of the infrastructure provider which must test various matching rules for service discovery.
-
8/9/2019 Testing in Service-Oriented Environments
24/79
14 | CMU/SEI-2010-TR-011
Infrastructural
Capability
Implication for SOA Testing Recommendation for SOA Testing
Service Usage
Monitoring
Service usage data is a source of information
for service providers in identifying who will be
affected by changes to a service. Monitoring
provides data important for audits and planning
(e.g., security audits, planning for increases inload).
Rec. 15: Develop a testing strategy for ser-
vice monitoring capabilities.
Service Virtualization Service virtualization creates and manages
virtual endpoints for web services, allowing the
service to be dynamically associated with physi-
cal endpoints in scenarios such as load-
balancing.
Rec. 16: Develop a testing strategy for ser-
vice virtualization features.
4.2 Testing Web Service Functionality
Services provide the main functionality or capability in an SOA environment. An initial step in
testing is to determine whether each service meets its functional requirements. While testing the
functionality of SOA services uses standard component testing methods, it poses a greater chal-
lenge because of the lack of control over many aspects of the test situation, including the services,SOA infrastructure, and other components with which the service being tested must interact. This
section highlights approaches to address the challenges of testing service functionality.10
The use of stubs to mock the behavior of unavailable components is a common software testing
strategy. This method has applied to the testing of services that need to interact with other services
where not all of the dependent elements are in place. Wolff, Godage, and Lublinsky [22,23,24]
show how mocking can be used to test services. Both open source (e.g., Apache Synapse11
) and
commercial implementations (e.g., SOAPSimulator, soapUI) can help service developers to create
mock services for unit testing quickly because not all functionality needs to be tested.
However, this approach has certain limitations including the following:
The quality of the testing result depends on the fidelity of the mock service to the actual ser-
vice. Services that behave in a complex manner are difficult to mock well, because under-
standing the complex behaviors and implementing these behaviors with testing tools is itself
a complex task. This leads to practical limitations of the types of services that can be
mocked.
The behaviors and interfaces of services to be mocked are often not finalized since develop-
ment is not complete: this is why the services must be mocked. This complication inevitably
leads to uncertainty in the reliability of testing results.
QoS cannot be adequately tested if services or other environment components are mocked.
10 Appendix C contains lists of attributes for testing functionality and other aspects of web services.
11For more on Synapse, go to http://synapse.apache.org/.
http://synapse.apache.org/http://synapse.apache.org/ -
8/9/2019 Testing in Service-Oriented Environments
25/79
15 | CMU/SEI-2010-TR-011
Recommendations for SOA Testing
Rec. 17: Provide support for service mocking, including mocking of services on remote platforms and
infrastructures.
Rec. 18: Integrate service mocking into the development and testing strategies.
4.3 Fault Injection
Fault injection, when used as a standard software testing approach, makes deliberate errors in the
implementation to see if the system can detect the fault and recover from it. When applied to ser-
vice testing, fault injection focuses on determining that services dont produce unintended beha-
viors or go into unreachable states when unexpected or faulty inputs are provided. This technique
can be applied to both individual web services [25] and composites [26].
Fuzzing is a testing technique supported by many tools that generates invalid data inputs to web
services. The tools capture the results of the fuzz and support the analysis of results to deter-
mine whether it compromised the web service.
Examples of types of faults that can be injected are a malformed service interface [27] (e.g., bad
WSDL definition), perturbation in the message [25] (e.g., null values in a SOAP message), and
timing delays in a composite (e.g., a service may timeout if it doesnt get a response). Fault injec-
tion is also used to test the security, reliability, and performance of a service as discussed in Sec-
tions 6 and 7.
Recommendation for SOA Testing
Rec. 19: Develop guidelines for the type of faults that services and service compositions must be
tested for.
A standard testing practice is testing the software to determine whether it handles realistic data
appropriately; such testing applies equally to services. In order to access appropriate test data, the
SLA should document the responsibilities for producing test data sets and appropriate test inter-
faces. As always, care must be taken to obfuscate production data since web services undergoing
testing may not assure the same degree of security as production services, nor are the persons per-
forming the testing guaranteed to be authorized to access production data.
Recommendation for SOA Testing
Rec. 20: Formal agreements should be in place with data providers to identify test data sets and testinterfaces.4.4 Regression Testing
Because of the rapid evolution of an SOA environment, testing cannot be a one-time event. Ser-
vices, service composites, end-to-end threadsand applications using themcontinue to evolve.
This section focuses on the specific aspects of regression testing that need to be considered in an
-
8/9/2019 Testing in Service-Oriented Environments
26/79
16 | CMU/SEI-2010-TR-011
SOA environment. Regression testing involves testing and identifying bug fixes to ensure that
defects have not been introduced as a result of evolution. Regression testing also attempts to de-
velop some level of assurance that things that were not touched did not break.
Regression testing in SOA environments is challenging because every service composition, busi-
ness process, and other system using a particular web service is affected by changes made to con-
stituent service [21], unlike a component-based context where the new release of a component
affects only the new versions of that component.
A service baseline aids regression testing efforts by making it easier to determine whether new
functionality has introduced defects and to more quickly mitigate the effects of those defects.
Testing tools that provide regression testing capabilities for web services offer the ability to take a
service baseline snapshot, so that subsequent changes can be automatically detected and reported.
Changes in that baseline can serve as triggers that indicate the need to perform regression testing
of end-to end mission threads. Several of these triggers are
upgrades/version changes to requestor/responder application, infrastructure, services in-
voked, back-end systems
upgrades to Service Contracts (WSDL) or SLAs (e.g., changes to expectations for the servic-
es within the mission thread)
changes or revisions to life-cycle management of components
changes in deployment configurations
changes to back-end systems or data sources
changes in a services functional behavior orQoS
updates to the rules used by testing tools
deprecation or retirement of a service
An SOA test sandbox is an infrastructure that accurately mirrors the deployment environment. A
test sandbox not only makes regression testing easier, but it also allows the use of automated test-
ing tools that can perform regression testing as soon as new or updated services are deployed after
a successful build process. Service developers can review the results of the automated testing
tools and initiate mitigation where appropriate.
The actual determination of whether to perform regression testing will be based on many factors,
including the nature of changes and the profile of the mission thread. Clearly, an organization will
be more risk-averse with, and therefore more likely to perform regression testing on threads that
can affect human life or cause widespread environmental or economic damage.
-
8/9/2019 Testing in Service-Oriented Environments
27/79
17 | CMU/SEI-2010-TR-011
Recommendations for SOA Testing
Rec. 21: Establish baselines for services, compositions, and end-to-end threads.
Rec. 22: Establish regression testing triggers for each.
Rec. 23: Establish a regression testing sandbox fitted with appropriate testing and analysis tools to
automate regression testing to the degree possible.
4.5 Testing Business Processes That Span Traditional Boundaries
End-to-end testing in an SOA environment refers to testing of the entire pathway involved in ex-
ecuting a specific mission or business process. For one example, in a military application, Time
Sensitive Targeting (TST) is often considered to have six phases: find, fix, track, target, engage,
and assess. These activities may span a number of application and organization boundaries. They
cover the point from which a potential target is identified to the point when it is no longer consi-
dered a threat. End-to-end testing of this capability in an SOA environment would involve execut-
ing a number of services, applications, and input from humans. The invoked SOA services may be
provided by different U.S. military services and other agencies, and possibly by coalition forces.
End-to-end testing needs to consider:
Decentralized ownership and lack of centralized control: Distributed ownership in an SOA
environment often involves setting up appropriate data and process context in back-end sys-
tems that are outside the control of the end-to-end tester. Also, services may be created by
different service providers, hosted on different infrastructures, and used by a variety of con-
sumers. These complications introduce the need for creating an environment (through nego-
tiation) for testing, testing across common infrastructures, testing across disparate infrastruc-
tures, and coordinating testing with multiple service providers and owners of back-end
systems.
Long-running business activities: SOA implementations often support the business processes
and workflow that constitute long-running business activities. For example, a loan approval
business process could be executed in phases over weeks.
Loosely Coupled Elements: Web services deployed in an SOA environment are loosely
coupled; they make minimal assumptions about the sending and receiving parties, so that
changes in one entity will have minimal impact on the other interfacing entities. The advan-
tage of such loose coupling is that testing becomes an issue of testing the composition
through simple interactions rather than complex ones.
Complexity: An end-to-end implementation is usually an execution pathway though multiple
people, services, composites, and infrastructure capabilities. Informationabout specific
services invoked, parameter data passed, and attachments includedis embedded in proto-
cols that are intended to support machine-to-machine interaction.
Regression testing: Changes at any service, node, or component along the pathway exercised
in support of an end-to-end thread may indicate a need for regression testing of the thread.
Maintaining awareness of these changes requires agreements regarding what types of
changes require notification, when such changes are allowed to occur, and how affected par-
ties are notified. An approach to bounding the need for regression testing is that of assurance
-
8/9/2019 Testing in Service-Oriented Environments
28/79
18 | CMU/SEI-2010-TR-011
cases (see Section 9.6) which provide a documented chain of reasoning based on evidence to
support a claim. In this case assurance cases could be used to help identify the conditions
under which further regression testing would be required.
Recommendations for SOA Testing
Rec. 24: Use approaches such as assurance cases to identify when additional regression testing is
needed.
Rec. 25: Develop a strategy for testing at each of the layers of the SOA interoperation stack.
Rec. 26: Instrument test instances to help identify the sources of failures.
Rec. 27: Perform regression testing to identify the impacts of changes on services, nodes and com-
ponents along the pathway that is exercised.
4.5.1 Process for End-to-End Testing
The basic process for testing end-to-end threads in an SOA environment does not differ from thehigh-level steps involved in other testing processes, with one exception: not only must tests be
devised to establish the correctness of predefined mission threads, but additional tests must be
devised to address the potential for dynamically composed mission threads (see Figure 1). Some
details of the steps are unique to SOA environments, as indicated in subsequent sections.
-
8/9/2019 Testing in Service-Oriented Environments
29/79
19 | CMU/SEI-2010-TR-011
Figure 1: Mission Thread Testing Process
4.5.2 Defining End-to-End Threads and Establishing Test Scenarios
With SOA as with any other technology that has many possible threads of controlcritical
threads must be identified and specifically tested for both functional and quality characteristics.
Several general approaches are used to identify end-to-end threads and establish test scenarios for
SOA environments. Manual approaches analyze mission functions, decompose these functions
into key decision steps and associated conditions, and define and build test cases accordingly. The
following process steps, derived from Sikri [28] are representative of manual processes:
1. Build a tree of high-level functions including inputs, outputs, and execution paths.
2. Identify conditions (e.g., timing, environment, and data) for each function in the tree.
3. Establish test scenarios based on functions and conditions.
4. Build tests cases addressing functions, conditions, and scenarios.
Other techniques are often needed to understand quality of service attributes and relate them to
various scenarios. A technique such as the Carnegie Mellon
Software Engineering Institute (SEI)
Mission Thread Workshop12
elicits quality attribute needs associated with the end-to-end threads
Carnegie Mellon is registered in the U.S. Patent and Trademark Office by Carnegie Mellon University.
12For more information about the Mission Thread Workshop, go tohttp://www.sei.cmu.edu/architecture/tools/evaluatingsos/mission.cfm.
http://www.sei.cmu.edu/architecture/tools/evaluatingsos/mission.cfmhttp://www.sei.cmu.edu/architecture/tools/evaluatingsos/mission.cfm -
8/9/2019 Testing in Service-Oriented Environments
30/79
20 | CMU/SEI-2010-TR-011
of a systems-of-systems environment [29]. However, for certain types of services (e.g., public
web services on the Internet) where the actual operating context (e.g., expected load and usage,
level of security required, etc.) is unknown until the service is actually being used, it may be ne-
cessary to adopt an incremental approach in which functionality is fielded and the system is then
adjusted to meet QoS goals.
Automated approaches have recently been developed that model mission threads in formal lan-
guages or that capture mission threads and establish test cases from existing modeling language
(e.g., Unified Modeling Language [UML]) artifacts. The resulting structures are suitable for many
types of analysis (e.g., dependency analysis, consistency checking, and test case generation)
[30,31,32]. While these approaches have significant potential for supporting rapid analysis of dy-
namic mission threads, the modeling language techniques are still in the maturing stage, so using
them needs to be done in small settings and with careful attention to results.
Recommendations for SOA Testing
Rec. 28: Develop a strategy for identifying the important threads in an end-to-end threads and deco-
rating those threads with both functional and QoS expectations.
Rec. 29: Develop test cases for critical threads that have been identified.
Rec. 30 For certain types of services (e.g., public web services on the internet) where the actual op-
erating context (e.g., expected load and usage, level of security required, etc.) is unknown
until the service is actually being used, it may be necessary to adopt an incremental ap-
proach in which functionality is fielded and the system is then adjusted to meet QoS goals.
4.5.3 Identifying Dynamic Mission Threads
We believe that at this stage of SOA maturity, the use of dynamically composed mission threads
in SOA environments has been oversold. Dynamic composition requires alignment at several lay-
ers of the interoperability stack (see Figure 2 for details) for the services and applications invoked.Requisite standards supporting such alignment are not yet available. As a result, dynamic compo-
sition in current SOA environments will rely on local conventions and standards that are likely to
be inconsistently defined and hard to enforce across organizations.
Dynamic composition is more likely among services and applications that are recognized as re-
lated and for which local conventions and standards have been established and tested. We recom-
mend that for SOA environments where dynamic composition is anticipated, every effort should
be made to identify and test likely compositions. Such testing is necessary to identify not only
successful operation of the dynamic thread, but also the effects of dynamic threads on network,
SOA infrastructure, individual services, and other critical threads.
Recommendations for SOA Testing
Rec. 31: Develop a strategy for identifying and testing likely dynamically composed threads.
Rec. 32: Implement non-interference strategies such that other threads cannot compromise function-
ing or QoS in the rest of the end-to-end thread.
-
8/9/2019 Testing in Service-Oriented Environments
31/79
21 | CMU/SEI-2010-TR-011
4.5.4 Identifying and Initializing the Context and Test Payloads
Establishing the testing context for end-to-end thread tests is a well-known problem in traditional
testing and continues to be a problem for SOA testing. It involves realistic simulation of the envi-
ronment in terms of capabilities, data, and loading. This section focuses on the specific aspects of
context to consider in end-to-end testing in SOA environments.
SOA environments often depend on a specific data value or other response from a black box ser-
vice. Well-understood principles that need to be continued are to (1) start with agreements with
service providers concerning the type of test data available, as well as rules governing testing of
the services, (2) use tools to simulate the production environment, (3) use data sets with high fi-
delity to enable testing such QoS attributes as performance, and (4) establish realistic loads on
infrastructure components (network and SOA) as well as on the invoked services and related ap-
plications.
Recommendation for SOA Testing
Rec. 33: Develop a strategy for cross-site and cross-organizational collaboration to identify test con-
text and data loads.
4.5.5 Capturing and Analyzing Results
End-to-end processing requires that the collaborating services, infrastructures, and applications
not only execute an action as directed but also that the action performs the correct activity. Cap-
turing data regarding the performance of services invoked in an end-to-end test may require in-
strumentation of the services or infrastructures on which they execute. The format of this metric
data differs according to the characteristics of the infrastructure and tools employed. These chal-
lenges can be addressed by making specific decisions on the data that is to be captured, by provid-
ing a logging service and by using a specially instrumented infrastructure.
An automated test tool offers value by collating and consolidating the test results to provide a sin-
gle point of review, making it easier to identify, categorize, and publish defects.13
Testing tools
that are a part of a comprehensive design and development suite offer the added benefit of linking
test plans to requirements, ensuring that the notion of requirements traceability is satisfied.
13In addition, a recent report from the Aberdeen Group [62] suggests that automated tool support for testing inSOA environments is a critical differentiator between the most sophist icated employers of SOA (best in class)and those that are less sophisticated. Only the related practice of through-life quality managementwas cited bymore bestin-class organizations as being critical to successful SOA.
Recommendation for SOA Testing
Rec. 34: Develop a strategy for cross-site and cross-organizational verification of changes in back-end
systems and capturing and analysis of test results.
-
8/9/2019 Testing in Service-Oriented Environments
32/79
22 | CMU/SEI-2010-TR-011
-
8/9/2019 Testing in Service-Oriented Environments
33/79
23 | CMU/SEI-2010-TR-011
5 Testing For Interoperability
A fundamental assumption of SOA adoption is that the resulting collection of services and infra-
structures will provide the basis for the creation of new end-user applications or end-to-endthreads through reuse of existing services. A consequence of this assumption is the reliance on the
interoperation of services that werent specifically designed to interoperate. Interoperation of the
individual components in any distributed environment (in this case, an SOA environment) relies
on agreements at several levels between service requestors and service providers, which makes
the execution of mission threads difficult. A further complication is that interoperation is the inte-
raction between two entities whereas testing traditionally focuses on entities and not their interac-
tions.
5.1 Defining Interoperability Levels
To achieve end-to-end testing, the services have to be interoperable. Figure 2 identifies the types
of agreements that must exist between service consumers and service providers to ensure success-
ful interoperation.
Figure 2: SOA Interoperation Stack Showing Levels of Interoperability
-
8/9/2019 Testing in Service-Oriented Environments
34/79
24 | CMU/SEI-2010-TR-011
Agreements need to be made between service providers and service consumers at all levels. The
agreements include the following (reading Figure 2 from the bottom up):
Agreements at the Machine levelProtocols for network-to-network and machine-to-
machine interaction, such that -two heterogeneous systems can exchange data irrespective of
their hardware, system-level software, and network characteristics. For example, an applica-
tion running on an Army tactical network can gain access to and interact with a service sys-tem running on an Army enterprise-wide network. This level of interoperability calls for
physical, data link, networking, transport, and session layers of the Open Systems Intercon-
nection Basic Reference Model (OSIRM).14
Agreements at the Syntactic levelAlignment of the structure and syntax of data and control
structures and interactions of applications and services. Within the web services domain, the
major standards include WSDL, SOAP, and, more recently, REST.
Agreements at the Semantic levelSemantic meaning at the application level on data and
control interactions among services. These agreements may reflect shared ontologies across
the interacting components (services, back end systems, etc.) or mechanisms for map-
ping/translating between different semantic models. Technologies used by web services toestablish common semantics for data include general purpose languages like the RDF and
specialized vocabularies like. OWL, OWL-S,15
and WSMO13
have been used to define com-
puter-interpretable definitions of service semantics. The same concept can apply to control
models. For example, for services implementing transactions to successfully interact, there
must be shared semantics for transactions (such as WS-AtomicTransaction or WS-
Coordination) or a mapping across individual transaction models. Unfortunately, technolo-
gies that provide semantic consistency are not as well developed, stable, or accepted as
agreements at the lower levels of the SOA interoperability stack.
Agreements at the Organizational levelCharacteristics of predefined mission threads or
business processes, including the associated context and workflows of thread participants, as
well as the appropriate qualities of service required. Some of the necessary agreements at
this level are defined in SLAs. WS-BPEL also provides a language to define an executable
process that involves message exchanges with other systems, such that the message ex-
change sequences are controlled by the WS-BPEL designer [33].
Agreements at the Situational levelCharacteristics of specific situation in which mission
threads or business processes are being applied. These characteristics are always unique to a
specific context. For example, consider a soldier calling in air support. While there are exist-
ing processes that are executed when calling in air support, the specifics of the urgency with
which air support is dispatched and how it is delivered will depend on the proximity of the
enemy, air assets available, and many other factors. There are no public standards for this
specific contextual information, but there may be less formal semantics and heuristics that
are applied.
14Acronyms used in this report are identified in Appendix A.
15 Formerly DAML-S; OWL-S is the Ontology Web Language for Services.
-
8/9/2019 Testing in Service-Oriented Environments
35/79
25 | CMU/SEI-2010-TR-011
Only the lowest two layers of the stack (machine and syntactic alignment) are addressed by com-
monly used web service standards. However, the standards that exist allow for testing services in
isolation for their conformance to those standards.
5.2 Interoperability in Infrastructure
The SOA infrastructure provides components for the lower two layers of the interoperability stack
and, as such, is most amenable to testing for conformance to well-defined standards such as those
provided by W3C. Such conformance testing is discussed later (see Section 8); however, there is
other infrastructure testing to be performed. Specifically, it is important to test that the compo-
nents within any one infrastructure interoperate appropriately and, where needed, that there is ap-
propriate interoperability between infrastructures.
5.2.1 Testing Within the Infrastructure
Typically the SOA infrastructure will be composed of one or more COTS products that have been
brought together to form a coherent infrastructure satisfying the organizations needs. The prod-
ucts should be selected through an evaluation process that enables the developers to select theproducts most suited to satisfy the requirements. The test community can participate in the evalua-
tion process using their expertise in testing as the mechanism for evaluating the products. Such
evaluation should consider not only the features of the individual products but also how well the
products fit with the rest of the candidate or already-selected products. Once the infrastructure has
been assembled, testing can be used to develop confidence that the infrastructure meets the needs.
We have already discussed testing the functionality of the infrastructure (See Section 3.1). How-
ever, additional testing could include testing for other quality attributes as well as simple assur-
ance that the infrastructure provides the basic connectivity between service consumers and pro-
viders.
5.2.2 Testing Between Infrastructures
Increasingly, SOA infrastructures are joined with other SOA infrastructures; it becomes essential
to identify these configurations and to test to ensure that they federate as required. Some federa-
tion configuration types are
homogenous infrastructure: This is essentially a horizontal scaling of the SOA infrastructure
where the federated infrastructures are different instances from the same base. Testing this
federation primarily involves checking connectors between the instances. For example, if the
same ESB product is used by all infrastructure nodes, the main testing task is to ensure that
the connection between two instances of an ESB by the same vendor functions as desired.
heterogeneous infrastructure: This configuration contains nodes that are different
sometimes the difference may only be in terms of versions; it is more common than the ho-mogeneous type because nodes are often planned, implemented, and upgraded separately.
hybrid infrastructure: In this configuration, some nodes are identical and others are different;
testing will require tactics from both homogeneous and heterogeneous infrastructure testing.
Regardless of the nature of the federation, it is important to perform the testing that would be ap-
plied to a single infrastructure, but across the federation in order to provide confidence that the
federation behaves as a single virtual infrastructure. Capabilities of particular importance are
-
8/9/2019 Testing in Service-Oriented Environments
36/79
26 | CMU/SEI-2010-TR-011
federated identity management and discovery. Without appropriate interoperation between the
infrastructures in at least these areas, the federation will not achieve the organizations aims.
Recommendation for SOA Testing
Rec. 35: Develop a strategy to identify and test the capabilities of SOA infrastructures that are needed
to federate the infrastructure.
5.3 Interoperability in Services
Services are found at both layers two and three of the interoperability stack(Figure 2), with infra-
structure services tending towards being in layer two and services providing business functionali-
ty tending towards layer three. Regardless of the source of the services (infrastructure or busi-
ness), testing them for interoperability can be handled in the same way, through means of
syntactic, semantic, and end-to-end interoperability testing. It should be noted that interoperability
is a property that exists between two services and is not a property inherent in the services them-
selves; thus testing services for interoperability becomes largely a matter of testing the service
interfaces adherence to standards.
When web services are being used, testing for syntactic interoperability can be performed by de-
termining that the services conform to the appropriate standards (see Chapter 8). Even when web
services are not being used, the SOA infrastructure will have a defined collection of protocols that
it supports and syntactic testing is an issue of determining that the service adheres to those proto-
cols.
Layer three describes semantic interoperability between services, which poses a problem for test-
ing the meaning of service interoperation. In order for services to be semantically interoperable,
they must perform according to expectations, thus a basis for interoperability is that each service
meets its specification, such testing has already been discussed in Section 4.2.
A requirement, though, for two services to be able to interoperate semantically is that they agree
on the data to be communicated. To the extent that the data models are well described, either in
common ontologies or in ontologies that can be mapped to each other, the test community can test
the potential for interoperation by ensuring that the service interface adheres to the defined mod-
el(s). Clearly this isnt sufficient for full interoperability testing; however, even such minimal
checking will be of benefit to the service developers.
Recommendations for SOA Testing
Rec. 36: Develop a list of interoperability standards that services must comply with, and consider of-
fering an interoperability compliance check as a service to developers.
Rec. 37: Develop a list of data models that services must use and consider offering a compliance
check as a service to developers.
5.4 End-to-End Interoperability
End-to-end interoperability may be found at the top two levels of the interoperability stack
(Figure 2). At level four, two or more organizations align their business processes with each other
and an end-to-end thread may cross the organizational boundaries, invoking services owned and
-
8/9/2019 Testing in Service-Oriented Environments
37/79
27 | CMU/SEI-2010-TR-011
operated by each other. Level five of the stack allows for the development of end-to-end threads
in a dynamic fashion in order to meet a specific demand. No matter how the thread arises, the is-
sues for testing are the same: to ensure that a sequence of discrete steps (the services) can be in-
voked even though they are geographically separate and owned by different organization and that
the invocation leads to desirable behavior.
Given that the number of services is potentially large, the number of possible interactions grows
exponentially, and not all interactions could have meaning, it is clearly not possible to test all inte-
ractions to ensure that the services are semantically interoperable. Testing interactions should be
largely performed in some context, specifically that of known endto-end threads. Composite ser-
vices, such as those defined in, say, Business Process Execution Language (BPEL), are a special
case of end-to-end threads. The interfaces to composite services can be tested in the same way
that services are tested, and the interactions between the services within the composite can be
tested in the same way that end-to-end threads are tested.
As a result, it is difficult to ensure consistency at the semantic and organization levels.16
The
most common strategy for achieving consistency at these levels is to develop predefined mission
threads (i.e., static threads) and test to ensure that consistency is achieved in use of vocabulary
and organizational interactions. Consistency at the situational level will often involve human-to-
human communication of the nuances of specific contexts and therefore it is difficult (or more
likely impossible) to make entirely consistent.
To address these challenges, a set of guidelines can be followed that includes
thoroughly test the services using the techniques defined in Section 4.2
test adherence to data standards
functionally test the defined threads
test the threads for performance, security, and other quality attributes
Failure at any of level in the stack can lead to a failure of mission thread. However, up to this
point (i.e., for predefined mission threads) an SOA environment does not necessarily differ signif-
icantly from other distributed environments. It is possible to tailor individual capabilities and ser-
vices, back-end systems, and network and SOA infrastructure to support the needs of specific
mission threads. For example, some organizations are tailoring the SOA deployment architecture
by adding additional hardware (e.g., hardware accelerators for XML parsers) or standing up addi-
tional SOA infrastructure. Taking these kinds of steps enables SOA end-to-end testing to be con-
ducted in ways similar to testing in any distributed environment.
A big selling point of SOA has been the ability to dynamically compose services in response to
specific, emerging situations. Scenarios are commonly presented where personnel check a registry
to determine what services are available and quickly craft and deploy a unique solution perfectlymatched to an emerging problem. However, dynamic composition (and late binding in general)
provides an exceptional challenge for verifying interoperability. The activities to verify syntactic,
semantic, and organizational interoperability must be performed on the fly against the specific
scenario presented by the dynamic composition. At best, only testing will only show that each
16While semantic alignment is being addressed by additional web service standards (e.g., OWL-S, WS-Coordination), these standards are not widely used.
-
8/9/2019 Testing in Service-Oriented Environments
38/79
28 | CMU/SEI-2010-TR-011
service conforms to the standards in the system including, ideally, to a common vocabulary (or
vocabularies that can be mapped to each other).
Standardized mechanisms to support the dynamic composition of the semantic, organizational,
and situational aspects of interactions are either immature or simply not available. Therefore, any
dynamically created end-to-end threads must be analyzed to ensure that they do not become error
prone and subject to inconsistencies. Those threads will not have been previously considered in
the testing process, and there is little assurance that they interoperate beyond the very basic syn-
tactic level provided by a WSDL specification. It is also useful to capture dynamically created
scenarios for future testing and/or incorporation into future releases.
We recommend that at this point of SOA maturity, compositions should be predefined and tested
for interoperability as well as other quality attributes. Mechanisms should be put in place to as-
sure, to the extent possible, that these predetermined compositions are not compromised by dy-
namic, on-the-fly compositions of services. Dynamic compositions, if allowed, will likely be use
at your own risk for the near term.
Recommendations for SOA Testing
Rec. 38: Develop a strategy to supports scenario-driven testing of interoperability.
Rec. 39: Develop policies for differential treatment of predefined and dynamic compositions of servic-
es.
Rec. 40: Develop policies for capturing dynamically created threads for future analysis and inclusion
into the collection of predefined threads.
-
8/9/2019 Testing in Service-Oriented Environments
39/79
29 | CMU/SEI-2010-TR-011
6 Testing for Security
SOA implementations are susceptible to many of the same security threats that afflict other net-
worked systems. Common vulnerabilities have been found in SOA implementations (both infra-structure and SOA services), including
denial-of-service attacks on SOAP message handlers and XML parsers
attacks that bypass authentication and authorization mechanisms or spoof identities
attacks that capture usernames and passwords
attacks that execute malicious scripts (often unknown to a benign pawn)
For more information about these and similar attacks, search the Department of Homeland Securi-
tys National Vulnerability Database [34] for SOA, XML, SOAP, and other common SOA terms.
6.1 Thread Modeling and Attack Surface
Maintaining a secure SOA infrastructure and services involves far more than testing and includes
appropriate development and deployment practices, security monitoring during operations, and
maintenance of the security profile during sustainment. However, the focus here on testing for
security must start with models of the threats facing SOA implementations and potential targets of
those threatsthat is, on threat modeling and attack surface.
6.1.1 Threat Modeling
Threat Modeling is a technique that is often used to generate security requirements and assess
possible security countermeasures, but it also provides a starting point for security testing. The
STRIDE Threat Model [35] provides six high-level threat categories to identify potential sources
of security failure in systems. The six threat categories in STRIDE, along with an example of an
SOA threat, include:
Spoofing: a malicious party taking on the identity of a legitimate web service requester, such
as by submitting different faked credentials.
Tampering: a malicious party altering the content of data and/or messages, for example by
uploading incorrect data. Tampering attacks normally happen in the context of other forms of
attacks such as spoofing that provide inappropriate access.
Repudiation: a party denying the initiation or authoring of an event or transaction. Repudia-
tion can occur when SOA transactions are not correctly logged.
InformationDisclosure: actively gaining access to supposedly secure information, for exam-
ple through lack of (or insufficient) encryption of SOAP messages.
DenialofService: limiting or eliminating access to a system or resource. Common denial of
service vectors in SOA include SOAP and XML parsers.
ElevationofPrivilege: taking on higher privileges than appropriate, such as a normal user
taking on super user or administrative privileges. Again, problems with SOAP message han-
dling have led to such attacks.
-
8/9/2019 Testing in Service-Oriented Environments
40/79
30 | CMU/SEI-2010-TR-011
The purpose of a Threat Model is to provide decision support for selecting and locating counter-
measures in the system. Categories of countermeasures to specific sorts of threats are identified in
the first two columns ofTable 4.
Table 4: Types of Threats and Typical Countermeasures
Threat Countermeasure Standard
Spoofing Authentication WS-Security
Tampering Signature WS-Security + XML Signature
Repudiation Audit Logging None
Information Disclosure Encryption, Input Validation Encryption: WS-Security + XML Encryption
Input Validation: None
Denial of Service Availability None
Elevation of Privilege Authorization None
The third column ofTable 4 identifies common standards that can apply the countermeasures.
Where no widely adopted and implemented industry standard is available, the wordNone is used
in the table. This does not imply that no standards existjust that the standard is not widely im-
plemented and adopted, as in the case of the XACML[36] standard for authorization. Many coun-
termeasures can be implemented in various ways with differing degrees of security provided (i.e.,
there is significant flexibility in th